Apple Silicon standalone build cannot connect to mlagents-learn

[TV4Fun]

Describe the bug The standard checkout of ml-agents includes prebuilt OSX x64 libraries that can't be used by an Apple Silicon build. For example com.unity.ml-agents/Plugins/ProtoBuffer/runtimes/osx/native/libgrpc_csharp_ext.x64.bundle. This can cause some strange behavior if you're trying to train using a standalone executable on Apple Silicon.

To Reproduce Steps to reproduce the behavior:

1. Checkout ml-agents from GitHub and build
2. Attach your ml-agents build to a unity project
3. Build the project as a standalone Apple Silicon executable
4. Run mlagents-learn --env=<PATH_TO_APP>

Console logs / stack traces Player-0.log:

Mono path[0] = '/Users/jcroteau/New Unity Project/Standalone.app/Contents/Resources/Data/Managed'
Mono config path = '/Users/jcroteau/New Unity Project/Standalone.app/Contents/MonoBleedingEdge/etc'
Found 1 interfaces on host : 0) 192.168.254.27
Multi-casting "[IP] 192.168.254.27 [Port] 55195 [Flags] 2 [Guid] 2162196733 [EditorId] 1012424985 [Version] 1048832 [Id] OSXPlayer(1,Nomad.home) [Debug] 1 [PackageName] OSXPlayer [ProjectName] New Unity Project" to [225.0.0.222:54997]...
Starting managed debugger on port 56733
Using monoOptions --debugger-agent=transport=dt_socket,embedding=1,server=y,suspend=n,address=0.0.0.0:56733
Initialize engine version: 2021.2.0b4 (af9ec38f7da3)
[Subsystems] Discovering subsystems at path /Users/jcroteau/New Unity Project/Standalone.app/Contents/Resources/Data/UnitySubsystems
GfxDevice: creating device client; threaded=1; jobified=0
 preferred device: Apple M1 (high power)
Metal devices available: 1
0: Apple M1 (high power)
Using device Apple M1 (high power)
Initializing Metal device caps: Apple M1
Begin MonoManager ReloadAssembly
- Completed reload, in  0.045 seconds
Metal RecreateSurface[0x11240f150]: surface size 2880x1800
UnloadTime: 3.480250 ms
Unexpected exception when trying to initialize communication: System.IO.IOException: Error loading native library "/Users/jcroteau/New Unity Project/Standalone.app/Contents/Resources/Data/Managed/../../../Plugins/libgrpc_csharp_ext.x64.bundle"
  at Grpc.Core.Internal.UnmanagedLibrary..ctor (System.String[] libraryPathAlternatives) [0x00063] in <2f154ad39ec14cfea604815989d96352>:0 
  at Grpc.Core.Internal.NativeExtension.Load () [0x000d7] in <2f154ad39ec14cfea604815989d96352>:0 
  at Grpc.Core.Internal.NativeExtension..ctor () [0x00006] in <2f154ad39ec14cfea604815989d96352>:0 
  at Grpc.Core.Internal.NativeExtension.Get () [0x00022] in <2f154ad39ec14cfea604815989d96352>:0 
  at Grpc.Core.Internal.NativeMethods.Get () [0x00000] in <2f154ad39ec14cfea604815989d96352>:0 
  at Grpc.Core.GrpcEnvironment.GrpcNativeInit () [0x00000] in <2f154ad39ec14cfea604815989d96352>:0 
  at Grpc.Core.GrpcEnvironment..ctor () [0x0001e] in <2f154ad39ec14cfea604815989d96352>:0 
  at Grpc.Core.GrpcEnvironment.AddRef () [0x00028] in <2f154ad39ec14cfea604815989d96352>:0 
  at Grpc.Core.Channel..ctor (System.String target, Grpc.Core.ChannelCredentials credentials, System.Collections.Generic.IEnumerable`1[T] options) [0x00051] in <2f154ad39ec14cfea604815989d96352>:0 
  at Grpc.Core.Channel..ctor (System.String target, Grpc.Core.ChannelCredentials credentials) [0x00000] in <2f154ad39ec14cfea604815989d96352>:0 
  at Unity.MLAgents.RpcCommunicator.Initialize (System.Int32 port, Unity.MLAgents.CommunicatorObjects.UnityOutputProto unityOutput, Unity.MLAgents.CommunicatorObjects.UnityInputProto& unityInput) [0x00008] in /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Communicator/RpcCommunicator.cs:220 
  at Unity.MLAgents.RpcCommunicator.Initialize (Unity.MLAgents.CommunicatorInitParameters initParameters, Unity.MLAgents.UnityRLInitParameters& initParametersOut) [0x00041] in /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Communicator/RpcCommunicator.cs:111 
UnityEngine.StackTraceUtility:ExtractStackTrace () (at /Users/bokken/buildslave/unity/build/Runtime/Export/Scripting/StackTrace.cs:37)
UnityEngine.DebugLogHandler:LogFormat (UnityEngine.LogType,UnityEngine.Object,string,object[])
UnityEngine.Logger:Log (UnityEngine.LogType,object)
UnityEngine.Debug:Log (object)
Unity.MLAgents.RpcCommunicator:Initialize (Unity.MLAgents.CommunicatorInitParameters,Unity.MLAgents.UnityRLInitParameters&) (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Communicator/RpcCommunicator.cs:140)
Unity.MLAgents.Academy:InitializeEnvironment () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:445)
Unity.MLAgents.Academy:LazyInitialize () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:279)
Unity.MLAgents.Academy:.ctor () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:248)
Unity.MLAgents.Academy/<>c:<.cctor>b__82_0 () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:117)
System.Lazy`1<Unity.MLAgents.Academy>:ViaFactory (System.Threading.LazyThreadSafetyMode)
System.Lazy`1<Unity.MLAgents.Academy>:ExecutionAndPublication (System.LazyHelper,bool)
System.Lazy`1<Unity.MLAgents.Academy>:CreateValue ()
System.Lazy`1<Unity.MLAgents.Academy>:get_Value ()
Unity.MLAgents.Academy:get_Instance () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:132)
Unity.MLAgents.DecisionRequester:Awake () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/DecisionRequester.cs:57)

(Filename: /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Communicator/RpcCommunicator.cs Line: 140)

Couldn't connect to trainer on port 5005 using API version 1.5.0. Will perform inference instead.
UnityEngine.StackTraceUtility:ExtractStackTrace () (at /Users/bokken/buildslave/unity/build/Runtime/Export/Scripting/StackTrace.cs:37)
UnityEngine.DebugLogHandler:LogFormat (UnityEngine.LogType,UnityEngine.Object,string,object[])
UnityEngine.Logger:Log (UnityEngine.LogType,object)
UnityEngine.Debug:Log (object)
Unity.MLAgents.Academy:InitializeEnvironment () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:459)
Unity.MLAgents.Academy:LazyInitialize () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:279)
Unity.MLAgents.Academy:.ctor () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:248)
Unity.MLAgents.Academy/<>c:<.cctor>b__82_0 () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:117)
System.Lazy`1<Unity.MLAgents.Academy>:ViaFactory (System.Threading.LazyThreadSafetyMode)
System.Lazy`1<Unity.MLAgents.Academy>:ExecutionAndPublication (System.LazyHelper,bool)
System.Lazy`1<Unity.MLAgents.Academy>:CreateValue ()
System.Lazy`1<Unity.MLAgents.Academy>:get_Value ()
Unity.MLAgents.Academy:get_Instance () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs:132)
Unity.MLAgents.DecisionRequester:Awake () (at /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/DecisionRequester.cs:57)

(Filename: /Users/jcroteau/ml-agents/com.unity.ml-agents/Runtime/Academy.cs Line: 459)

Output of mlagents-learn --env=Standalone.app --debug:

                        ▄▄▄▓▓▓▓
                   ╓▓▓▓▓▓▓█▓▓▓▓▓
              ,▄▄▄m▀▀▀'  ,▓▓▓▀▓▓▄                           ▓▓▓  ▓▓▌
            ▄▓▓▓▀'      ▄▓▓▀  ▓▓▓      ▄▄     ▄▄ ,▄▄ ▄▄▄▄   ,▄▄ ▄▓▓▌▄ ▄▄▄    ,▄▄
          ▄▓▓▓▀        ▄▓▓▀   ▐▓▓▌     ▓▓▌   ▐▓▓ ▐▓▓▓▀▀▀▓▓▌ ▓▓▓ ▀▓▓▌▀ ^▓▓▌  ╒▓▓▌
        ▄▓▓▓▓▓▄▄▄▄▄▄▄▄▓▓▓      ▓▀      ▓▓▌   ▐▓▓ ▐▓▓    ▓▓▓ ▓▓▓  ▓▓▌   ▐▓▓▄ ▓▓▌
        ▀▓▓▓▓▀▀▀▀▀▀▀▀▀▀▓▓▄     ▓▓      ▓▓▌   ▐▓▓ ▐▓▓    ▓▓▓ ▓▓▓  ▓▓▌    ▐▓▓▐▓▓
          ^█▓▓▓        ▀▓▓▄   ▐▓▓▌     ▓▓▓▓▄▓▓▓▓ ▐▓▓    ▓▓▓ ▓▓▓  ▓▓▓▄    ▓▓▓▓`
            '▀▓▓▓▄      ^▓▓▓  ▓▓▓       └▀▀▀▀ ▀▀ ^▀▀    `▀▀ `▀▀   '▀▀    ▐▓▓▌
               ▀▀▀▀▓▄▄▄   ▓▓▓▓▓▓,                                      ▓▓▓▓▀
                   `▀█▓▓▓▓▓▓▓▓▓▌
                        ¬`▀▀▀█▓

 Version information:
  ml-agents: 0.28.0.dev0,
  ml-agents-envs: 0.28.0.dev0,
  Communicator API: 1.5.0,
  PyTorch: 1.10.0a0+gite856a45
2021-07-24 13:29:24 DEBUG [learn.py:220] Configuration for this run:
2021-07-24 13:29:24 DEBUG [learn.py:221] {
    "default_settings": null,
    "behaviors": {
        "KinematicMoveToGoal": {
            "trainer_type": "ppo",
            "hyperparameters": {
                "batch_size": 1000,
                "buffer_size": 10000,
                "learning_rate": 0.0003,
                "beta": 0.005,
                "epsilon": 0.2,
                "lambd": 0.95,
                "num_epoch": 5,
                "learning_rate_schedule": "linear"
            },
            "network_settings": {
                "normalize": false,
                "hidden_units": 256,
                "num_layers": 2,
                "vis_encode_type": "simple",
                "memory": null,
                "goal_conditioning_type": "hyper"
            },
            "reward_signals": {
                "extrinsic": {
                    "gamma": 0.95,
                    "strength": 1.0,
                    "network_settings": {
                        "normalize": false,
                        "hidden_units": 128,
                        "num_layers": 2,
                        "vis_encode_type": "simple",
                        "memory": null,
                        "goal_conditioning_type": "hyper"
                    }
                }
            },
            "init_path": null,
            "keep_checkpoints": 5,
            "checkpoint_interval": 500000,
            "max_steps": 5000000,
            "time_horizon": 64,
            "summary_freq": 10000,
            "threaded": true,
            "self_play": null,
            "behavioral_cloning": null
        }
    },
    "env_settings": {
        "env_path": "Standalone.app",
        "env_args": null,
        "base_port": 5005,
        "num_envs": 1,
        "seed": -1
    },
    "engine_settings": {
        "width": 84,
        "height": 84,
        "quality_level": 5,
        "time_scale": 20,
        "target_frame_rate": -1,
        "capture_frame_rate": 60,
        "no_graphics": false
    },
    "environment_parameters": null,
    "checkpoint_settings": {
        "run_id": "KinematicBiggerPenalties",
        "initialize_from": null,
        "load_model": false,
        "resume": true,
        "force": false,
        "train_model": false,
        "inference": false,
        "results_dir": "results"
    },
    "torch_settings": {
        "device": null
    },
    "debug": true
}
2021-07-24 13:29:24 DEBUG [learn.py:245] run_seed set to 5998
2021-07-24 13:29:24 DEBUG [torch.py:58] default Torch device: cpu
2021-07-24 13:29:24 DEBUG [stats_writer.py:60] Initializing StatsWriter plugins: default
2021-07-24 13:29:24 DEBUG [stats_writer.py:63] Found 3 StatsWriters for plugin default
2021-07-24 13:29:24 DEBUG [env_utils.py:33] The true file name is Standalone
2021-07-24 13:29:24 DEBUG [env_utils.py:105] The launch string is /Users/jcroteau/New Unity Project/Standalone.app/Contents/MacOS/New Unity Project
2021-07-24 13:29:24 DEBUG [env_utils.py:106] Running with args ['--mlagents-port', '5005', '-logFile', '/Users/jcroteau/New Unity Project/results/KinematicBiggerPenalties/run_logs/Player-0.log']
[UnityMemory] Configuration Parameters - Can be set up in boot.config
    "memorysetup-bucket-allocator-granularity=16"
    "memorysetup-bucket-allocator-bucket-count=8"
    "memorysetup-bucket-allocator-block-size=4194304"
    "memorysetup-bucket-allocator-block-count=1"
    "memorysetup-main-allocator-block-size=16777216"
    "memorysetup-thread-allocator-block-size=16777216"
    "memorysetup-gfx-main-allocator-block-size=16777216"
    "memorysetup-gfx-thread-allocator-block-size=16777216"
    "memorysetup-cache-allocator-block-size=4194304"
    "memorysetup-typetree-allocator-block-size=2097152"
    "memorysetup-profiler-bucket-allocator-granularity=16"
    "memorysetup-profiler-bucket-allocator-bucket-count=8"
    "memorysetup-profiler-bucket-allocator-block-size=4194304"
    "memorysetup-profiler-bucket-allocator-block-count=1"
    "memorysetup-profiler-allocator-block-size=16777216"
    "memorysetup-profiler-editor-allocator-block-size=1048576"
    "memorysetup-temp-allocator-size-main=4194304"
    "memorysetup-job-temp-allocator-block-size=2097152"
    "memorysetup-job-temp-allocator-block-size-background=1048576"
    "memorysetup-job-temp-allocator-reduction-small-platforms=262144"
    "memorysetup-temp-allocator-size-background-worker=32768"
    "memorysetup-temp-allocator-size-job-worker=262144"
    "memorysetup-temp-allocator-size-preload-manager=262144"
    "memorysetup-temp-allocator-size-nav-mesh-worker=65536"
    "memorysetup-temp-allocator-size-audio-worker=65536"
    "memorysetup-temp-allocator-size-cloud-worker=32768"
    "memorysetup-temp-allocator-size-gfx=262144"
2021-07-24 13:30:24 WARNING [environment.py:431] Environment timed out shutting down. Killing...
2021-07-24 13:30:24 DEBUG [subprocess_env_manager.py:220] UnityEnvironment worker 0: environment stopping.
2021-07-24 13:30:24 DEBUG [subprocess_env_manager.py:234] UnityEnvironment worker 0 closing.
2021-07-24 13:30:24 DEBUG [subprocess_env_manager.py:237] UnityEnvironment worker 0 done.
2021-07-24 13:30:24 DEBUG [trainer_controller.py:81] Saved Model
2021-07-24 13:30:24 DEBUG [subprocess_env_manager.py:368] SubprocessEnvManager closing.
2021-07-24 13:30:24 DEBUG [subprocess_env_manager.py:107] UnityEnvWorker 0 got exception trying to close.
Traceback (most recent call last):
  File "/Users/jcroteau/miniforge3/envs/pytorch-metal/bin/mlagents-learn", line 33, in <module>
    sys.exit(load_entry_point('mlagents', 'console_scripts', 'mlagents-learn')())
  File "/Users/jcroteau/ml-agents/ml-agents/mlagents/trainers/learn.py", line 250, in main
    run_cli(parse_command_line())
  File "/Users/jcroteau/ml-agents/ml-agents/mlagents/trainers/learn.py", line 246, in run_cli
    run_training(run_seed, options)
  File "/Users/jcroteau/ml-agents/ml-agents/mlagents/trainers/learn.py", line 125, in run_training
    tc.start_learning(env_manager)
  File "/Users/jcroteau/ml-agents/ml-agents-envs/mlagents_envs/timers.py", line 305, in wrapped
    return func(*args, **kwargs)
  File "/Users/jcroteau/ml-agents/ml-agents/mlagents/trainers/trainer_controller.py", line 173, in start_learning
    self._reset_env(env_manager)
  File "/Users/jcroteau/ml-agents/ml-agents-envs/mlagents_envs/timers.py", line 305, in wrapped
    return func(*args, **kwargs)
  File "/Users/jcroteau/ml-agents/ml-agents/mlagents/trainers/trainer_controller.py", line 105, in _reset_env
    env_manager.reset(config=new_config)
  File "/Users/jcroteau/ml-agents/ml-agents/mlagents/trainers/env_manager.py", line 68, in reset
    self.first_step_infos = self._reset_env(config)
  File "/Users/jcroteau/ml-agents/ml-agents/mlagents/trainers/subprocess_env_manager.py", line 333, in _reset_env
    ew.previous_step = EnvironmentStep(ew.recv().payload, ew.worker_id, {}, {})
  File "/Users/jcroteau/ml-agents/ml-agents/mlagents/trainers/subprocess_env_manager.py", line 98, in recv
    raise env_exception
mlagents_envs.exception.UnityTimeOutException: The Unity environment took too long to respond. Make sure that :
     The environment does not need user interaction to launch
     The Agents' Behavior Parameters > Behavior Type is set to "Default"
     The environment and the Python interface have compatible versions.
     If you're running on a headless server without graphics support, turn off display by either passing --no-graphics option or build your Unity executable as server build.

Environment (please complete the following information):

• Unity Version: 2021.2.0b4
• OS + version: macOS 11.5
• ML-Agents version: latest main branch from source
• Torch version: 1.10.0a0+gite856a45 (latest main built from source)

[TV4Fun]

Note: This also happens when running inside the Apple Silicon build of Unity Editor. The only way I'm able to train is to use the Intel build.

[TV4Fun]

Trying to get this to work. The default gRPC Unity plugin doesn't include an Arm64 build, but I was able to make my own. The Unity editor and player isn't finding the native library to include with the app, but if I build a standalone and manually add the Apple Silicon dylib I built to it, then it works and has a pretty substantial performance improvement. If anyone can help me in getting the editor to find the right Arm64 file, I would greatly appreciate it.

[ervteng]

cc: @surfnerd who might have some experience with Apple Silicon

[surfnerd]

I have a PR I am working on that adds the arm64 arch to the native plugin. It should be ready this week.

[surfnerd]

See #5283

[TV4Fun]

Confirmed that this now works correctly with Unity Arm64 on the latest main branch, so closing this issue.

[github-actions[bot]]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.